{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":15,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":15,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12","author_layer_release":"2026-06-26"},"query_hash":"8b186b870421","filters":{"venue":"Educational Assessment Evaluation and Accountability"}},"results":[{"id":"W2183435816","doi":"10.1007/s11092-015-9233-6","title":"Teacher assessment literacy: a review of international standards and measures","year":2015,"lang":"en","type":"review","venue":"Educational Assessment Evaluation and Accountability","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":253,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Literacy; Educational assessment; Mainland; Psychology; Pedagogy; Geography","authors":[{"name":"Christopher DeLuca","is_ca":true},{"name":"Danielle LaPointe-McEwan","is_ca":true},{"name":"Ulemu Luhanga","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1374248929728662,"gpt":0.5795117938372106,"spread":0.4420869008643444,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01989881,0.0003648247,0.001121853,0.0002033982,0.0002839256,0.0002925214,0.0004009207,0.0002018631,0.004369651],"category_scores_gemma":[0.001066191,0.0003140791,0.0002119145,0.000385426,0.000352463,0.0006630399,0.0001730027,0.0003782468,0.000004934088],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002266516,"about_ca_system_score_gemma":0.01331173,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002951071,"about_ca_topic_score_gemma":0.0001042388,"domain_scores_codex":[0.9920936,0.001714225,0.001078812,0.0006411524,0.00422735,0.0002448302],"domain_scores_gemma":[0.9944112,0.0006462519,0.0008778481,0.0003379707,0.003531978,0.0001946938],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000003666461,0.000416556,0.008758956,0.009534821,0.0002063789,8.375743e-8,0.001017851,8.103374e-8,9.133324e-8,0.003787145,0.01047102,0.9658033],"study_design_scores_gemma":[0.0002945763,0.0000268005,0.009968014,0.01069064,0.0008722066,0.000001745219,0.0008651337,0.00002604948,9.680338e-9,0.001504737,0.9754583,0.0002917486],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0000709102,0.9466183,0.00002833235,0.0029385,0.001037275,0.002312272,0.0002103592,0.00002144773,0.04676258],"genre_scores_gemma":[0.0005903608,0.9954776,0.001223664,0.0001656051,0.0007095266,0.0006238716,0.0007282111,0.00002123979,0.0004599099],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9655116,"threshold_uncertainty_score":0.9999312,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1983949565","doi":"10.1007/s11092-008-9063-x","title":"Using data to support educational improvement","year":2008,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":127,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; Ministry of Education and Child Care","funders":"","keywords":"Data collection; Context (archaeology); Computer science; Scale (ratio); Data science; Knowledge management; Sociology","authors":[{"name":"Carol Campbell","is_ca":true},{"name":"Ben Levin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5729116068459882,"gpt":0.5914743349389096,"spread":0.01856272809292137,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009523775,0.0002822558,0.000309444,0.0003668474,0.0007907636,0.0003798606,0.001150019,0.00006906506,0.01716572],"category_scores_gemma":[0.002532561,0.0002437913,0.00007075669,0.0008388679,0.0001724385,0.001773869,0.0006156723,0.0001842769,0.0002764126],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008636583,"about_ca_system_score_gemma":0.007289428,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002441478,"about_ca_topic_score_gemma":0.0001137195,"domain_scores_codex":[0.9923185,0.0003259483,0.001205736,0.001406622,0.004377117,0.0003661245],"domain_scores_gemma":[0.9938186,0.001460957,0.0004173559,0.001571877,0.002373523,0.0003577132],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003351561,0.002356694,0.8399504,0.00001828512,0.00007072533,1.725713e-7,0.001082572,0.0003208209,0.005371096,0.01199339,0.09438781,0.04441447],"study_design_scores_gemma":[0.0004879148,0.0001286852,0.9345005,0.000007925356,0.00004411086,0.00001434336,0.001179701,0.01157585,0.00006393185,0.03753353,0.01414955,0.0003139248],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9519051,0.00005192179,0.001440684,0.03811243,0.001870631,0.001457268,0.0001948144,0.00001774548,0.00494946],"genre_scores_gemma":[0.9707683,0.00000981536,0.02274924,0.002447888,0.0008855059,0.0002988775,0.0009421558,0.00001608073,0.001882176],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09455007,"threshold_uncertainty_score":0.9983383,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1948718713","doi":"10.1007/s11092-015-9231-8","title":"Towards a framework for the validation of early childhood assessment systems","year":2015,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":47,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"Fok Ying Tong Education Foundation; U.S. Department of Education","keywords":"Early childhood education; Early childhood; Accountability; Educational assessment; Conceptual framework; Argument (complex analysis); Field (mathematics); Work (physics); Psychology; Political science; Pedagogy; Developmental psychology; Sociology; Medicine; Social science; Engineering","authors":[{"name":"Jessica Goldstein","is_ca":false},{"name":"Jessica Kay Flake","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09283552875712486,"gpt":0.4581947012178403,"spread":0.3653591724607154,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008678145,0.0001239495,0.0001656455,0.00007297563,0.0005459117,0.0002702371,0.0002156873,0.00009055461,0.0003503435],"category_scores_gemma":[0.001567636,0.00009843498,0.00005909672,0.0003032168,0.0001776795,0.0004028367,0.00003316244,0.0001361631,0.000005212901],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006816606,"about_ca_system_score_gemma":0.01149056,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001700383,"about_ca_topic_score_gemma":0.00007317229,"domain_scores_codex":[0.9971274,0.0004937285,0.0004484478,0.000293064,0.001444737,0.0001926245],"domain_scores_gemma":[0.9962591,0.00115191,0.0003077649,0.00025855,0.001869753,0.0001529063],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00001544795,0.0007594113,0.3011886,0.00003166212,0.0000937679,6.650843e-9,0.04654329,0.0004117904,0.000005081875,0.6267889,0.002498144,0.021664],"study_design_scores_gemma":[0.0003740203,0.00004655649,0.8803624,0.00002453547,0.00006080298,2.968199e-7,0.02218005,0.000444361,0.0000077376,0.09221762,0.004166219,0.0001153915],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9351956,0.0005047503,0.01719308,0.03079848,0.003383263,0.003622845,0.00003933405,0.00003809184,0.009224563],"genre_scores_gemma":[0.984349,0.00003698677,0.01378398,0.0001868039,0.0006503231,0.0007753148,0.00008951315,0.0000085235,0.0001195919],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5791739,"threshold_uncertainty_score":0.9941134,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2000242929","doi":"10.1007/s11092-014-9195-0","title":"Navigating dilemmas in transforming assessment practices: experiences of mathematics teachers in Ontario, Canada","year":2014,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":34,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Manitoba; University of Ottawa","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Situated; Accountability; Judgement; Pedagogy; Assessment for learning; Curriculum; Best practice; Mathematics education; Faculty development; Professional development; Psychology; Formative assessment; Political science; Computer science","authors":[{"name":"Christine Suurtamm","is_ca":true},{"name":"Martha Koch","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0639113308409666,"gpt":0.4511814897350099,"spread":0.3872701588940433,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.007483715,0.0001323771,0.0002364628,0.00004948715,0.0002141798,0.00009123502,0.0001902647,0.00006220774,0.001675645],"category_scores_gemma":[0.0006279103,0.0001323064,0.00003027302,0.0002912103,0.0001409721,0.0006708196,0.0000261183,0.000335228,4.239731e-7],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002408778,"about_ca_system_score_gemma":0.007218331,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9363547,"about_ca_topic_score_gemma":0.9939944,"domain_scores_codex":[0.9968973,0.0005048651,0.0006190287,0.0003094209,0.001427922,0.0002414273],"domain_scores_gemma":[0.9979545,0.001110368,0.0004901019,0.0001431114,0.0002327548,0.00006909706],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000002206206,0.0003801483,0.8405382,0.00002033674,0.000006491381,4.639756e-8,0.151267,0.00004699335,0.00003669066,0.004677813,0.00002114941,0.003002951],"study_design_scores_gemma":[0.0003766728,0.00001571663,0.7005319,0.00006494693,0.00001344698,1.452413e-7,0.2946471,0.001841067,0.000006590236,0.001827118,0.0005634249,0.0001118859],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9820801,0.00002126454,0.00006413774,0.002352575,0.0003463716,0.0007457044,0.000001386395,0.000005854846,0.01438261],"genre_scores_gemma":[0.9953084,0.000008397139,0.004129766,0.00007143129,0.00006525064,0.0003286995,0.00003082077,0.000005344497,0.0000518942],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1433801,"threshold_uncertainty_score":0.9992369,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2043024755","doi":"10.1007/s11092-013-9184-8","title":"The journey from rhetoric to reality: participatory evaluation in a development context","year":2014,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Citizen journalism; Context (archaeology); Situated; Stakeholder; Sociology; Warrant; Rhetoric; Participatory action research; International development; Participatory evaluation; Engineering ethics; Conceptual framework; Epistemology; Political science; Social science; Public relations; Computer science; Engineering; Geography; Business","authors":[{"name":"Jill Anne Chouinard","is_ca":true},{"name":"J. Bradley Cousins","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3829268271012413,"gpt":0.5696211298539817,"spread":0.1866943027527404,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.06747412,0.0002122991,0.0002798938,0.0002317745,0.0006216108,0.0006665331,0.0004642842,0.00007634415,0.004233742],"category_scores_gemma":[0.008169542,0.0001473888,0.00005576446,0.0007548869,0.00009947614,0.0006588169,0.0001231143,0.0002194223,0.0002591008],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001369351,"about_ca_system_score_gemma":0.003834401,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003053016,"about_ca_topic_score_gemma":0.006621468,"domain_scores_codex":[0.9888198,0.002964711,0.001474645,0.0006977539,0.005707079,0.000336063],"domain_scores_gemma":[0.9911026,0.004836776,0.0004337771,0.0007621041,0.002640866,0.0002238673],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00004077028,0.000305767,0.4019195,0.000003549824,0.00001987661,1.409377e-8,0.004641605,0.001246196,0.00007665333,0.007170225,0.003815394,0.5807605],"study_design_scores_gemma":[0.0007141675,0.00003069195,0.8032506,0.00001741,0.00002356886,3.255211e-7,0.003767718,0.07966953,0.00003599855,0.05223899,0.06009216,0.0001588828],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9576222,0.0001667016,0.001211732,0.03237257,0.001207644,0.001470756,0.000009139568,0.00001415241,0.005925114],"genre_scores_gemma":[0.9957601,0.0000116917,0.0008188507,0.001509501,0.0002735485,0.00131592,0.00009406466,0.000009037915,0.0002072532],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5806016,"threshold_uncertainty_score":0.9966765,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4283698056","doi":"10.1007/s11092-022-09389-9","title":"Mapping the constellation of assessment discourses: a scoping review study on assessment competence, literacy, capability, and identity","year":2022,"lang":"en","type":"review","venue":"Educational Assessment Evaluation and Accountability","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":31,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University; Memorial University of Newfoundland","funders":"","keywords":"Competence (human resources); Operationalization; Construct (python library); Authentic assessment; Literacy; Rubric; Pedagogy; Identity (music); Psychology; Engineering ethics; Sociology; Epistemology; Social psychology; Computer science; Engineering","authors":[{"name":"Andrew Coombs","is_ca":true},{"name":"Christopher DeLuca","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1960247782782946,"gpt":0.5663204934521253,"spread":0.3702957151738308,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02644831,0.0005563343,0.001589927,0.0002997073,0.001848462,0.0006231524,0.0007120628,0.0001322349,0.004770985],"category_scores_gemma":[0.0005171326,0.0004339578,0.0003241553,0.001160552,0.0006946883,0.00105317,0.0004963881,0.0008005917,0.000003911251],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001955419,"about_ca_system_score_gemma":0.008670215,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006379624,"about_ca_topic_score_gemma":0.0004651463,"domain_scores_codex":[0.9849502,0.007576329,0.001880366,0.001143113,0.004055768,0.0003941864],"domain_scores_gemma":[0.991549,0.005041269,0.001760233,0.0008045073,0.0006936481,0.0001513147],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.000007314198,0.005064021,0.22775,0.03832293,0.0005448615,6.639996e-7,0.009985577,0.000009465075,4.66841e-7,0.02000534,0.0004576786,0.6978517],"study_design_scores_gemma":[0.001430285,0.0004628765,0.7079127,0.07314509,0.003811967,0.000007352503,0.0634001,0.0003166077,2.667551e-8,0.004932126,0.1431074,0.001473473],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.006654717,0.9474011,0.0000959048,0.002751955,0.00170517,0.02296511,0.000119422,0.00004714508,0.01825948],"genre_scores_gemma":[0.04627483,0.9500688,0.0004894137,0.0001894756,0.000323269,0.002133829,0.0004301275,0.00002918744,0.00006109835],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6963782,"threshold_uncertainty_score":0.9998112,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3044087203","doi":"10.1007/s11092-020-09329-5","title":"Is Canada really an education superpower? The impact of non-participation on results from PISA 2015","year":2020,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Youth Substance Use and School Attendance","field":"Social Sciences","cited_by":31,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"H2020 European Research Council","keywords":"Superpower; Political science; Inclusion (mineral); Scale (ratio); Test (biology); Sample (material); Population; Economic growth; Psychology; Geography; Mathematics education; Demography; Sociology; China; Social psychology; Economics","authors":[{"name":"Jake Anders","is_ca":false},{"name":"Silvan Has","is_ca":false},{"name":"John Jerrim","is_ca":false},{"name":"Nikki Shure","is_ca":false},{"name":"Laura Zieger","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07773513248884605,"gpt":0.4788059318311023,"spread":0.4010707993422563,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001767048,0.000114332,0.0001334045,0.0000223571,0.0003629722,0.00009176704,0.0001952867,0.00004840927,0.001235366],"category_scores_gemma":[0.0006322169,0.00008710341,0.00004422255,0.000257175,0.0001023336,0.0004861469,0.00001353361,0.0001328848,0.000006506091],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007392292,"about_ca_system_score_gemma":0.012741,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.6321313,"about_ca_topic_score_gemma":0.3527885,"domain_scores_codex":[0.9976058,0.0004437369,0.0003829511,0.0003533537,0.001061216,0.0001529535],"domain_scores_gemma":[0.9980096,0.0004451429,0.0002333623,0.0002629363,0.0008731692,0.0001758103],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003225907,0.001066476,0.8137749,0.00001422735,0.00007534285,5.029923e-8,0.07704487,0.0009529603,0.000537086,0.00299657,0.08162513,0.02158979],"study_design_scores_gemma":[0.0003048994,0.00006389511,0.9830563,0.00001362912,0.00002713132,2.755593e-8,0.008178005,0.004182991,0.00003286292,0.003103487,0.0009371904,0.00009959281],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9503996,0.00004351757,0.00002108826,0.04205965,0.0002891814,0.0005591038,0.0002452985,0.000007970687,0.006374614],"genre_scores_gemma":[0.9970747,0.00002087194,0.0002568468,0.001623858,0.0005063367,0.00007856431,0.0003885721,0.000005494207,0.00004475603],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2793428,"threshold_uncertainty_score":0.9996777,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W944434266","doi":"10.1007/s11092-015-9224-7","title":"Juggling multiple accountability systems: how three principals manage these tensions in Ontario, Canada","year":2015,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":31,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"York University; Western University","funders":"","keywords":"Accountability; Negotiation; Mandate; Variety (cybernetics); Public relations; Work (physics); Public administration; Political science; Sociology; Engineering; Law","authors":[{"name":"Katina Pollock","is_ca":true},{"name":"Sue Winton","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2865166200978165,"gpt":0.4540437544114345,"spread":0.167527134313618,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01659711,0.0003323362,0.0004661094,0.0002741613,0.0002651783,0.0008296712,0.0005953899,0.000102623,0.001462609],"category_scores_gemma":[0.005165966,0.0002676469,0.00007566452,0.0006759102,0.000131463,0.001453105,0.0002400272,0.0004005366,0.00001965559],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.006059891,"about_ca_system_score_gemma":0.01753351,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.8374575,"about_ca_topic_score_gemma":0.9880023,"domain_scores_codex":[0.9917656,0.0006977224,0.001176315,0.001109423,0.004816791,0.0004342116],"domain_scores_gemma":[0.9918509,0.00304066,0.0005387631,0.000952447,0.003263129,0.0003541247],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002152015,0.0005128988,0.9897052,0.00001248329,0.00001899583,2.358355e-7,0.0008886769,0.001094104,0.00003720897,0.002048631,0.004154616,0.001505451],"study_design_scores_gemma":[0.0008406676,0.00003442592,0.9186205,0.00002515242,0.00002874164,0.000002577681,0.01274178,0.03055405,0.000005045766,0.02311672,0.01375048,0.0002799173],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.976284,0.0001570541,0.0002538894,0.0155245,0.002349422,0.001909127,0.0000524029,0.00001834938,0.003451269],"genre_scores_gemma":[0.9959942,0.000002620113,0.001256798,0.0004672329,0.0002203686,0.0007176913,0.0002051473,0.00001398428,0.001121926],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1505448,"threshold_uncertainty_score":0.9999776,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3188740232","doi":"10.1007/s11092-021-09368-6","title":"“I am here for the students”: principals’ perception of accountability amid work intensification","year":2021,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University; University of Manitoba; University of British Columbia","funders":"","keywords":"Accountability; Perception; Work (physics); Context (archaeology); Public relations; Psychology; Political science; Focus group; Task (project management); Pedagogy; Sociology; Management; Law","authors":[{"name":"Fei Wang","is_ca":true},{"name":"Cameron Hauseman","is_ca":true},{"name":"Katina Pollock","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2359750747103962,"gpt":0.5289763041093049,"spread":0.2930012293989087,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.007101198,0.0001324907,0.0001831941,0.00004347351,0.0007054121,0.0001788403,0.0002375204,0.00008528513,0.001734282],"category_scores_gemma":[0.002191524,0.0001111206,0.0000967566,0.0004099316,0.0004771691,0.0003376066,0.00005219273,0.0001588545,0.000007368194],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007097806,"about_ca_system_score_gemma":0.00187681,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005023057,"about_ca_topic_score_gemma":0.001090944,"domain_scores_codex":[0.997046,0.0007439008,0.0004584268,0.0004216959,0.001130741,0.0001992593],"domain_scores_gemma":[0.9949026,0.001613625,0.0002535675,0.0003641696,0.002797114,0.00006893422],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002483647,0.0008587654,0.9132749,0.00005085185,0.00005997618,4.852417e-9,0.05434141,0.00001674193,0.0001162564,0.007547404,0.001106228,0.02260256],"study_design_scores_gemma":[0.0002270834,0.000009088024,0.711866,0.00001446754,0.00007608449,1.800848e-7,0.2790726,0.00007901698,0.000009363783,0.002719924,0.005842066,0.00008411897],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9000383,0.0004638981,0.0005865459,0.0954692,0.000991618,0.001433624,0.00001671868,0.00002424571,0.000975861],"genre_scores_gemma":[0.996408,0.0001253634,0.001203966,0.0005057827,0.0004540435,0.0007395534,0.0001232763,0.000008060887,0.0004319286],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2247312,"threshold_uncertainty_score":0.9991783,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2040697633","doi":"10.1007/s11092-011-9117-3","title":"Consistency of report card grades and external assessments in a Canadian province","year":2011,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":21,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Trent University; University of Toronto","funders":"Ministère de l’Éducation, Gouvernement de l’Ontario","keywords":"Report card; Psychology; Consistency (knowledge bases); Reading (process); Medical education; Grade level; Variance (accounting); Mathematics education; Medicine; Pedagogy; Mathematics; Political science","authors":[{"name":"John A. Ross","is_ca":true},{"name":"Lynn Kostuch","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2160058372944961,"gpt":0.5025141811864667,"spread":0.2865083438919707,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009342493,0.0001706397,0.0002825066,0.0004062161,0.0001592485,0.0001403287,0.0002504874,0.00006430442,0.001696792],"category_scores_gemma":[0.001176474,0.0001457244,0.00004921446,0.0003820764,0.0002227257,0.0007954681,0.00007602631,0.0001438439,0.000005375001],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005645155,"about_ca_system_score_gemma":0.007392032,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.08166616,"about_ca_topic_score_gemma":0.09856468,"domain_scores_codex":[0.995932,0.0003557408,0.001071249,0.0006503976,0.00172735,0.0002632492],"domain_scores_gemma":[0.9970948,0.0006007866,0.000468155,0.0004224507,0.001174137,0.0002397363],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000009294571,0.0003574523,0.9734425,0.00001433155,0.00001147376,6.843753e-7,0.0005256787,0.000001388412,0.0001928558,0.01390222,0.0001823596,0.01135979],"study_design_scores_gemma":[0.0003273549,0.00006852171,0.898832,0.00002333757,0.0000225805,0.00001262335,0.002236972,0.0009575937,0.00006236765,0.09699287,0.0003325962,0.000131237],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9810081,0.0001813513,0.00008878965,0.00220075,0.0004796952,0.000840222,0.00002696683,0.000004825616,0.01516935],"genre_scores_gemma":[0.9936734,0.00001462146,0.005470767,0.0001742941,0.0000596811,0.0001849013,0.00004496698,0.000006483035,0.0003708813],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08309064,"threshold_uncertainty_score":0.9992158,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2076487748","doi":"10.1007/s11092-014-9200-7","title":"Personality and student performance on evaluation methods used in business administration courses","year":2014,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Management and Marketing Education","field":"Business, Management and Accounting","cited_by":14,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval; Université de Sherbrooke","funders":"","keywords":"Personality; Preference; Psychology; Sample (material); Multilevel model; Latent variable; Control (management); Applied psychology; Medical education; Social psychology; Computer science; Machine learning; Medicine; Artificial intelligence; Statistics; Mathematics","authors":[{"name":"Sawsen Lakhal","is_ca":true},{"name":"Serge Sévigny","is_ca":true},{"name":"Éric Frénette","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06513401286280288,"gpt":0.446581531766921,"spread":0.3814475189041181,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01504163,0.0002242176,0.0002104504,0.0003016889,0.0002863647,0.0004485783,0.0001267556,0.0000691586,0.0009954715],"category_scores_gemma":[0.001089798,0.0002176972,0.0000282801,0.0004787976,0.00008756985,0.001297743,0.00006971425,0.000143339,0.00001515302],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003219263,"about_ca_system_score_gemma":0.0003352833,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002343936,"about_ca_topic_score_gemma":0.0003161548,"domain_scores_codex":[0.9972629,0.0003872079,0.0004489969,0.0005871262,0.001109517,0.0002042353],"domain_scores_gemma":[0.9979853,0.0004767288,0.0003169525,0.0002684096,0.0009282555,0.0000243466],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005415414,0.0007273735,0.8874009,0.0002416023,0.00001449515,1.410801e-8,0.0001747126,0.0002255089,0.0000762847,0.02105574,0.0004483941,0.08958077],"study_design_scores_gemma":[0.000878557,0.0000228558,0.9096548,0.00005496459,0.0001005937,4.601344e-7,0.0006158612,0.07890879,0.000006861623,0.007065797,0.002477766,0.0002126693],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9696934,0.00003893174,0.0002292983,0.009403312,0.0007499734,0.001249859,0.000001359419,0.00003199061,0.01860183],"genre_scores_gemma":[0.9965197,0.00001676392,0.001043154,0.0008187844,0.0007036331,0.0005065812,0.0003025856,0.00001355973,0.00007520477],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0893681,"threshold_uncertainty_score":0.9999177,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2088210129","doi":"10.1007/s11092-013-9181-y","title":"Teacher supervision practices and characteristics of in-school supervisors in Uganda","year":2014,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":14,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval","funders":"","keywords":"Supervisor; Perception; Professional development; Psychology; Directive; Descriptive statistics; Medical education; Pedagogy; Medicine; Political science","authors":[{"name":"Lawrence Kalule","is_ca":true},{"name":"Yamina Bouchamma","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1368831847854667,"gpt":0.4800151790265387,"spread":0.343131994241072,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.006613798,0.00008800685,0.0001660281,0.0001221135,0.0001123937,0.00006489215,0.00007968751,0.00006468379,0.001177145],"category_scores_gemma":[0.003912318,0.00008615899,0.00001657024,0.000219379,0.0001852409,0.0004595655,0.00002440072,0.0001641066,0.000003168707],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002387773,"about_ca_system_score_gemma":0.0007627914,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003289933,"about_ca_topic_score_gemma":0.005179109,"domain_scores_codex":[0.9978762,0.0008931857,0.0003451729,0.0002567642,0.0004913215,0.0001373482],"domain_scores_gemma":[0.9985533,0.0007821831,0.0002120472,0.0001171074,0.0002656991,0.00006970895],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000007350328,0.0003556031,0.9472538,0.00002927805,0.000003479975,5.949698e-9,0.03878725,8.988889e-7,0.00003282025,0.001964292,0.00006079096,0.01150446],"study_design_scores_gemma":[0.0003187179,0.00001472207,0.9052666,0.00002309423,0.00001099832,1.215591e-7,0.08958882,0.0003726921,0.000001870568,0.002568704,0.001754008,0.00007957889],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9512207,0.0001631786,0.000005988679,0.04516311,0.0001995729,0.0003614718,0.000002230121,0.00000615359,0.002877625],"genre_scores_gemma":[0.9989753,0.0001260281,0.0003185036,0.0002231881,0.0001748011,0.00007962661,0.00002344948,0.000004109285,0.00007493604],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05080157,"threshold_uncertainty_score":0.9997359,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2531903003","doi":"10.1007/s11092-016-9247-8","title":"Two teacher quality measures and the role of context: evidence from Chile","year":2016,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"School Choice and Performance","field":"Social Sciences","cited_by":11,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; Institute for Christian Studies","funders":"","keywords":"Context (archaeology); Quality (philosophy); Teacher quality; Psychology; Geography; Political science; Business; Archaeology; Epistemology; Philosophy; Marketing","authors":[{"name":"María Verónica Santelices","is_ca":false},{"name":"Edgar Valencia","is_ca":true},{"name":"Jorge González","is_ca":false},{"name":"Sandy Taut","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09023241200854629,"gpt":0.4715839134556266,"spread":0.3813515014470803,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.008321559,0.00006852349,0.0001221981,0.00001826937,0.0003230932,0.0000645763,0.0001262083,0.00003388255,0.002225778],"category_scores_gemma":[0.00196466,0.00003912556,0.00003039537,0.00007947321,0.0005418906,0.0006025459,0.00002813703,0.00006793405,0.000006354024],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001329401,"about_ca_system_score_gemma":0.0008321269,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01559795,"about_ca_topic_score_gemma":0.007764817,"domain_scores_codex":[0.9976555,0.0009842711,0.0002532217,0.0002104189,0.0008000966,0.00009645605],"domain_scores_gemma":[0.9968752,0.002260586,0.0001623088,0.000178187,0.0004729685,0.00005071356],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004082567,0.00005737047,0.9004488,0.000002796343,0.0000120002,8.875521e-10,0.003989514,6.146504e-7,0.0003480156,0.01321801,0.0001102706,0.08177181],"study_design_scores_gemma":[0.0006471403,0.000005868363,0.9376786,0.00003353847,0.00002663999,5.39634e-8,0.0057693,0.0001285557,0.00006238864,0.05316682,0.002421165,0.00005987547],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9710794,0.002095249,0.00003375409,0.02191842,0.000152341,0.0003830745,0.00001641332,0.000007361477,0.004313952],"genre_scores_gemma":[0.9988016,0.0002513962,0.0001457087,0.0001729973,0.0004016503,0.0001038156,0.000005625815,0.000002278547,0.0001149603],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08171193,"threshold_uncertainty_score":0.9986863,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1566806182","doi":"10.1007/s11092-011-9132-4","title":"Learning as identity and practice through involvement in online moderation","year":2011,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Online and Blended Learning","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Queen's University","keywords":"Moderation; Context (archaeology); Psychology; Identity (music); Social psychology; Qualitative research; Element (criminal law); Pedagogy; Sociology; Political science; Social science","authors":[{"name":"Lenore Adie","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1552906905599422,"gpt":0.507281857070541,"spread":0.3519911665105987,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.005582265,0.00008334107,0.00009454074,0.00005715161,0.0004350377,0.0001343297,0.00006267048,0.0000585022,0.0012534],"category_scores_gemma":[0.00204731,0.00008691743,0.00001604467,0.0001973094,0.0001144641,0.00256078,0.00003863724,0.0002131262,0.000006837724],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003252764,"about_ca_system_score_gemma":0.001479413,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006586135,"about_ca_topic_score_gemma":0.008281832,"domain_scores_codex":[0.9976226,0.0009960651,0.0002553427,0.0002871988,0.0006979289,0.000140829],"domain_scores_gemma":[0.9987881,0.0004633625,0.0001432619,0.00008194482,0.0004720745,0.0000513219],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005661136,0.001908093,0.6853884,0.00003589662,0.00002651474,2.822758e-7,0.06113346,0.0002093572,0.0001086956,0.1945454,0.00009688237,0.05649045],"study_design_scores_gemma":[0.0004244139,0.00006353082,0.7667501,0.00001946075,0.00003936837,9.295689e-7,0.09831007,0.003124212,0.000005505015,0.1241806,0.006942419,0.0001394287],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.965082,0.0001294343,0.0002416857,0.0120824,0.0001716858,0.0004452803,0.000001925789,0.00001773051,0.02182788],"genre_scores_gemma":[0.9834878,0.0002413055,0.01531931,0.0004882682,0.0001762378,0.00006055484,0.0000801264,0.000004438414,0.0001419551],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08136173,"threshold_uncertainty_score":0.9996596,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4417293613","doi":"10.1007/s11092-025-09471-y","title":"Investigating the relationship between educational inequity and teacher participation in professional development: A cross-national and quasi-experimental approach using TIMSS","year":2025,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"Göteborgs Universitet; Vetenskapsrådet","keywords":"Socioeconomic status; Student achievement; Academic achievement; Professional development; Higher education; Social class","authors":[{"name":"Leah Natasha Glassow","is_ca":false},{"name":"Jan‐Eric Gustafsson","is_ca":false},{"name":"Nils Kirsten","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1803804953596685,"gpt":0.5374982171340111,"spread":0.3571177217743426,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.007677886,0.0001603094,0.0001519292,0.0001849634,0.001636193,0.0003357421,0.0001179169,0.0001145577,0.0002810466],"category_scores_gemma":[0.002693487,0.0001443274,0.00001964971,0.0004878996,0.0004877523,0.0007069399,0.00009571134,0.0002741914,0.00000292274],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001046116,"about_ca_system_score_gemma":0.008588704,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004698342,"about_ca_topic_score_gemma":0.0002601994,"domain_scores_codex":[0.996897,0.0009009916,0.0005634958,0.0004529207,0.0009594042,0.0002261223],"domain_scores_gemma":[0.9973276,0.001772045,0.0001983407,0.000110205,0.000473426,0.0001183572],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000003363568,0.000552112,0.9329837,0.00002509856,0.00001262946,2.548052e-9,0.01356413,0.0000319031,0.000007983896,0.05105527,0.000187485,0.00157633],"study_design_scores_gemma":[0.0004193579,0.000006175582,0.9670641,0.00004918331,0.00001650157,4.309672e-7,0.005094888,0.002274525,0.00000539398,0.02438518,0.0005369765,0.0001472285],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9836693,0.0002799493,0.0001510708,0.01080877,0.0004076905,0.0009405888,0.000005616555,0.00001404191,0.003723009],"genre_scores_gemma":[0.9933163,0.000003350821,0.004334368,0.0003850634,0.000236067,0.0004216654,0.0001876644,0.000006503455,0.00110902],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03408047,"threshold_uncertainty_score":0.9996635,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}